The doctor behind the first autonomous artificial intelligence-driven diagnostic system to win approval from the FDA is showing his work.
Michael Abramoff, a University of Iowa ophthalmologist and professor, and his colleagues spent more than three decades developing the IDx-DR, an AI-driven tool designed to diagnose degeneration in the eye.
The FDA earlier this year approved IDx-DR to detect retinopathy caused by diabetes, the first time such a technology was OK’d in any field in medicine. And now the researchers have published the clinical data behind IDx-DR’s success.
So how did they do it?
Abramoff tells FDAnews he began studying AI 30 years ago and a few years later started working on the algorithms that would become IDx-DR 22. He founded the company that now owns the marketing rights to IDx-DR in 2007 but was in discussions with the FDA for eight years before getting the go-ahead for a clinical trial last year.
The FDA designated IDx-DR as a breakthrough device and “provided intensive interaction and guidance to the company on efficient device development, to expedite development, to expedite evidence generation and the agency’s review of the device,” he says.
When the trial was finally greenlighted, researchers recruited 900 people with diabetes at risk for retinal decline at 10 eye clinics around the country. Over six months, clinic staff visually examined participants’ eyes, scanned them with IDx-DR and, finally, trained retinal photographers took pictures of their eyes using fundus imaging and optical coherence tomography, considered the “gold standard” for retinal imaging.
All but 80 volunteers completed all three procedures during the trial. Of those, 198 showed signs of retinal decay. The IDx-DR system diagnosed 173 correctly, a success rate of more than 87 percent.
Nearly 24,000 Americans suffer diabetes-related blindness each year, according to the CDC. But nearly half of diabetics in the U.S. skip recommended annual eye exams, the FDA noted in approving IDx-DR. Among the major reasons cited for skipping exams were cost and lack of access to advanced eye care — hurdles Abramoff believes IDx-DR can help overcome.
The IDx-DR has two algorithms at its core. One focuses on image quality and the other on diagnostics. The image algorithm takes four rapid-fire photos of the retina to determine whether there’s sufficient area, focus, color balance and exposure to make a diagnosis.
IDx-DR was developed specifically for diabetic retinopathy. But Abramoff believes it has potential to do a lot more. In fact, he and his team are working on sister devices that they hope will help diagnose other eye disorders such as glaucoma. — Bill Myers